Efficacy of microbial sampling recommendations and practices in sub-Saharan Africa

Abstract

Current guidelines for testing drinking water quality recommend that the sampling rate, which is the number of samples tested for fecal indicator bacteria (FIB) per year, increases as the population served by the drinking water system increases. However, in low-resource settings, prevalence of contamination tends to be higher, potentially requiring higher sampling rates and different statistical methods not addressed by current sampling recommendations. We analyzed 27,930 tests for FIB collected from 351 piped water systems in eight countries in sub-Saharan Africa to assess current sampling rates, observed contamination prevalences, and the ability of monitoring agencies to complete two common objectives of sampling programs: determine regulatory compliance and detect a change over time. Although FIB were never detected in samples from 75% of piped water systems, only 14% were sampled often enough to conclude with 90% confidence that the true contamination prevalence met an example guideline (≤5% chance of any sample positive for FIB). Similarly, after observing a ten percentage point increase in contaminated samples, 43% of PWS would still require more than a year before their monitoring agency could be confident that contamination had actually increased. We conclude that current sampling practices in these settings may provide insufficient information because they collect too few samples. We also conclude that current guidelines could be improved by specifying how to increase sampling after contamination has been detected. Our results suggest that future recommendations should explicitly consider the regulatory limit and desired confidence in results, and adapt when FIB is detected.

Keywords: Guidelines for drinking water quality; Microbial water quality; Sampling programs; Statistical uncertainty; Sub-saharan Africa; Water quality regulations.

Fig. 1

Classifying uncertainty about true the contamination prevalence and its relationship to the criteria for passing a regulatory limit of $\le 5\text{\%}$. CA: confidently above; UA: unconfidently above; UB: unconfidently below; CB: confidently below. Each category fails (F) or passes (P) the limit as shown in the lower grid.

Fig. 2

Annual equivalent sampling rates by PWS (points) compared with GDWQ recommendations (black line) and ± 50% of GDWQ recommendations (shaded grey) as tested by a) water suppliers (left) and b) surveillance agencies (right) by country (Benin (dots), Ethiopia (triangles), Kenya (squares), Uganda (crosses) and Zambia (crossed boxes)).

Fig. 3

Compliance of all PWS (points) in the dataset with an example regulatory limit specifying at most a 5% chance of samples containing FIB (red horizontal line). a) Required sampling rates to pass or fail the example limit at 75% (solid lines) and 90% (dashed lines) confidence. Y-axis ranges from 0 to 25%. b) The observed contamination prevalences for all 351 PWS in the dataset with error bars that span the range of values that the observed prevalence is above with 90% confidence and below with 90% confidence. Y-axis ranges from 0 to 100%. PWS are divided into four categories according to their adherence to the limit with 90% confidence: confidently above (more contaminated than) the limit (CA), confidently below the limit (CB), unconfidently above the limit (UA), and unconfidently below the limit (UB). (For interpretation of the references to colour in this figure legend, the reader is referred to the Web version of this article.)

Fig. 4

The proportion of PWS in the dataset (n = 351) that fall into each category of regulatory compliance: CA: confidently above; UA: unconfidently above; UB: unconfidently below; CB: confidently below. These categories map into failing (F) or passing (P) the regulatory limit differently for each of three criteria: benefit-of-the-doubt; face-value; and fail-safe.

Fig. 5

Sampling rate and ability to detect a change in water quality. a) Months until a decrease in water quality in a given PWS in the dataset could be concluded from a ten percentage point increase in the prevalence of FIB contamination with either 75% or 90% confidence. For each confidence level, the observable curve is created by PWS with extreme contamination prevalences (i.e. close to 0% or 100%). b) Distribution of the mean observed monthly sampling rates among all PWS.

Fig. 6

Statistical relationships for improving sampling recommendations. a) Harmonizing the number of samples and the regulatory limit for PWS with no (grey/lower lines) or one (black/upper lines) samples positive for FIB and with 75% (solid lines) or 90% (dashed lines) confidence. b) Increases in GDWQ recommended sampling rates required due to a fixed number of samples found to be positive for FIB, calculated for two population extremes: 5000 people (black/highest lines in each pair) and 5,000,000 people (grey/lower lines in each pair). The multiplier depends on the required confidence level: 75% (solid lines) or 90% (dashed lines).